Noise-Resilient Waveform Discrimination: Evaluating Kernel and Ensemble Methods Across Varying Signal-to- Noise Ratios

DOI

https://www.doi.org/10.59256/ijire.20260704022

Published In

Volume 7, Issue 4 (2026)

Published Online

July-August 2026

Pages

206-210

Article Type

Original Article

Check for updates

This paper inspects the time-domain and frequency-domain behavior of five types of simulated signals: sinusoidal, square, triangular, sawtooth, and chirp waves. The team built a dataset of 1,000 labelled samples, each mixed with different levels of additive white Gaussian noise. They extracted twelve carefully chosen features from every sample, covering things like root-mean-square amplitude, crest factor, spectral centroid, spectral roll-off, and zero-crossing rate. Then, they tested three supervised classifiers—Random Forest, Gradient Boosting, and a Radial-Basis-Function Support Vector Machine—using five-fold cross-validation. The SVM topped the charts, reaching a mean cross-validated accuracy of 99.60 ±0.37%, but all models scored above 97% accuracy, even with signal-to-noise ratios as low as 10 dB. Principal Component Analysis shows that the different signal types are clearly separated in the 12-dimensional feature space. The results indicate that a small set of spectro-statistical features is enough for robust signal classification, even across the noisy conditions typical in laboratory simulations.